Why re-uploads make fake videos harder to catch

A video that has travelled through three WhatsApp groups, been screen-recorded and re-posted to a Reel is not the same file it started as. Each step quietly throws information away, and some of that information is exactly what detectors use.

What compression does

Video files are huge, so every platform compresses them. Compression keeps what your eye notices, like shapes, colours and movement, and discards fine detail it thinks you won't miss. Each time a video is downloaded and re-uploaded, it's compressed again, and the losses add up.

What gets lost along the way

• Fine texture. The subtle pixel patterns that AI image generators leave behind are among the first things compression smooths away. Frame-based detectors depend on them.
• Resolution. Many apps shrink videos, sometimes to 480p or lower. Small faces and text lose the detail needed to spot errors.
• Frame rate and smooth motion. Some re-encodes drop frames or blend them, which disturbs the motion patterns that tell real cameras from generators.
• Metadata. Most platforms strip a file's metadata on upload, including camera details and Content Credentials that might have said "made with AI". A screen recording removes them completely.
• Audio quality. Voices get compressed too, hiding some of the artefacts that voice checks listen for.

Why this matters

A heavily compressed fake can look more convincing to people, because the blur hides glitches, and it also gives automated checks less to work with. Real videos suffer too: a compressed real clip can lose the natural detail that makes it look camera-made.

That's why a responsible detector should be trained and tested on compressed copies, not just pristine files. Ours was: we trained and tested our in-house model on every clip in several qualities, from the original down to heavily re-compressed versions, so its 95% balanced accuracy in testing includes degraded copies. It's still true that the better the copy, the more confident any check can be.

How to give a check the best copy

1. Find the original post. Use the earliest upload you can find. Our reverse video search guide shows how.
2. Paste the link rather than a forward. Video Verifier downloads the best quality available for the scan from Instagram, YouTube, TikTok, X and 1,000+ other sites. A Deep Scan fetches a sharper copy where it can.
3. Download, don't screen-record. If you have to upload a file, save the video itself rather than recording your screen.
4. Avoid trimmed and captioned copies. Re-posts with added text, borders or reaction cams have been re-encoded at least once more.

If a result comes back "too close to call", don't read it as a failure. It often means the copy you have simply doesn't hold enough evidence. A better copy can change that.

The bigger picture

Compression is one of the reasons no detector can promise certainty. Combined with new generators and real-plus-AI mixes, it's why we show the evidence and the limits behind every result. Read more in how AI video detectors work.